A Stability-Selected Transcriptomic Signature for Predicting Time-to-First-Treatment in Chronic Lymphocytic Leukemia
Materials and Methods/Case Presentation/Objective: Gene expression and clinical outcome data were obtained from the Gene Expression Omnibus. The training cohort included 130 patients with CLL from GSE39671, including 67 treatment events and 63 censored observations. An independent external validation cohort included 70 patients from GSE22762, including 35 treatment events and 35 censored observations. TTFT was defined as the interval from diagnosis to initiation of first therapy. Elastic Net Cox regression, nested cross-validation, and bootstrap stability selection were used to develop a leakage-free survival modeling framework. Preprocessing, feature ranking, and hyperparameter optimization were performed within training partitions. Bootstrap stability selection was conducted across 150 resampling iterations. The final model was evaluated using Harrell's concordance index, Kaplan-Meier analysis, hazard ratios, log-rank testing, and locked external validation without retraining or threshold adjustment. Pathway enrichment analysis was performed using Gene Ontology, KEGG, Reactome, and MSigDB Hallmark gene sets.
Results/Description/Main Outcome Measure(s):
Bootstrap stability selection identified a compact 30-gene transcriptomic signature associated with TTFT in CLL. The final model achieved a nested cross-validation concordance index of 0.6876 ± 0.0666 in the training cohort and maintained discrimination in the independent external validation cohort, with a concordance index of 0.7198. Risk stratification using the locked model and a prespecified training-derived threshold separated patients into distinct TTFT risk groups in the external cohort. High-risk patients had a significantly greater likelihood of requiring treatment than low-risk patients, with a hazard ratio of 3.95 (95% confidence interval, 1.63-9.57; log-rank p=0.0010). Pathway enrichment identified inflammatory response, TNF-α/NF-κB signaling, innate immune activation, KRAS-related signaling, and cellular responses to bacterial and lipid-derived stimuli as dominant biological themes.
Conclusion(s).: A stability-selected transcriptomic signature identified molecular patterns associated with earlier treatment requirement in CLL and retained discriminatory performance in an independent validation cohort. These findings support transcriptomic survival modeling as a complementary approach to TTFT risk stratification. Enrichment of inflammatory, innate immune, and NF-κB-related pathways suggests that immune- and microenvironment-associated biology may contribute to variation in treatment initiation timing. Further validation in larger, clinically annotated CLL cohorts incorporating established prognostic markers is warranted.
Trademarked Items:
References: 1. Molica S, Shanafelt TD, Giannarelli D, et al. The chronic lymphocytic leukemia international prognostic index predicts time to first treatment in early CLL: independent validation in a prospective cohort of early stage patients. Am J Hematol. 2016;91(11):1090-1095. doi:10.1002/ajh.24493.
2. Hallek M, Cheson BD, Catovsky D, et al. iwCLL guidelines for diagnosis, indications for treatment, response assessment, and supportive management of CLL. Blood. 2018;131(25):2745-2760. doi:10.1182/blood-2017-09-806398.
